Decomposition of bank loans and economic activity in TURKEY

ABSTRACT¹

¹ The views expressed in this paper are only those of the authors and should not be interpreted as reflecting those of the Central Bank of the Republic of Turkey. Email addresses: mailto:[email protected];[email protected];[email protected]

We examine the empirical link between loans and economic activity in Turkey with a focus on the components of loans by borrower (household/business) and by purpose of use (housing/personal) as well as currency of denomination (domestic/foreign). We estimate a separate VAR model for each type of loan and each GDP expenditure item to analyse whether different types of loans have different effects on economic activity and through what channels. According to our empirical results, credit shocks have statistically significant impact on economic activity, especially within the first two quarters. We find that shocks that expand household and TL-denominated business loans by the same rate have quite similar effects on private consumption, final domestic demand and GDP, while household loans has a much smaller impact on investment compared to business loans. While shocks to FX-denominated business loans have significant effect on total investment, they have much weaker effect on private consumption and GDP. The effect of housing loans on investment is found to be comparable to that of business loans, suggesting strong feedback between demand for housing and construction investment. We investigate the robustness of findings to alternative data samples, as well as some alternative identifying restrictions.

KEYWORDS:

• Household credit
• business credit
• GDP growth
• credit shocks

JEL CLASSIFICATION:

• E44
• E32
• C32.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

¹ The views expressed in this paper are only those of the authors and should not be interpreted as reflecting those of the Central Bank of the Republic of Turkey. Email addresses: mailto:[email protected];[email protected];[email protected]

² See Akçiğit et al. (2021) for firm-level effects of the Credit Guarantee Fund (CGF) program implemented in 2017 and Kara (2021) for the role of the policy-induced rise bank credit during the Covid-19 crisis.

³ Throughout the paper, household loans and consumer loans; and business loans and commercial loans are used interchangeably.

⁴ Housing loans are granted to real persons for the purpose of financing new or used housing purchases. It is a secured loan as the property is mortgaged to the lender as a security until the repayment of the loan is completed. On the other hand, personal loans are used by real persons to meet their individual needs and are generally unsecured loans.

⁵ Although the data for credit components go back to 2003, we restrict our baseline analysis to post-2009 due to two main concerns: i) the structural break in GDP and its components due to the methodological change in national accounts in December 2016, ii) more stable relationship between credit and growth after the global financial crisis. Nevertheless, we share estimation results for some of the models for a longer sample period in Section IV. Estimation results are more significant and robust for the post-2009 period.

⁶ See Büyükbaşaran, Karasoy-Can, and Küçük (2019) for identification of credit supply shocks in Turkey in a larger SVAR setup that analyse total private credit and GDP.

⁷ See for example Beck (2009), Rousseau and Wachtel (1998) and Luintel and Khan (1999)

⁸ In fact, VAR estimations with real interest rates instead of credit spread have worse diagnostics and give noisy, unreliable results.

⁹ Exchange rate adjustment is made by multiplying FX credit with the average value of the exchange rate basket (average of USD/TRY and EUR/TRY) between 2007 and 2011.

¹⁰ We omit vehicle loans deliberately, as banks’ share in this market is very low in comparison to non-bank institutions.

¹¹ Descriptive statistics of the data used in the estimations can be found in Appendix B.

¹² Details on the methodological change can be provided upon request.

¹³ We also follow the referee’s recommendation and calculate Jorda (2005) linear local projection IRFs as an alternative to standard VAR IRFs. The results show that the methodological change does not yield a remarkable change in estimation results in terms of the statistical significance of the responses. The paths of responses with local projections are a bit more volatile than the original responses, but the main trends are almost the same. Therefore, we assess that our results are also robust to estimation method. The results are not reported due to space limitations and are available upon request.

¹⁴ We report non-accumulated and accumulated response functions ofGDP and other expenditure items to 1 standard deviation shocks of each credit type in Appendix C. Responses of other endogenous variables to the shocks are not reported due to space limitations and are available upon request.

¹⁵ As an example, for the response of GDP to credit shock between time t and t + j, the formal representation is GDPt,t + j/CSt,t + j, where GDPt,t + j and CSt,t + j denote the accumulated changes in GDP and credit stock, respectively, in j periods of time.

¹⁶ Authors’ own calculation from 2012 Input-Output tables of Turkstat.

¹⁷ Table B1 in Appendix B reports the cross-correlations across different loan types to give some idea about the extent of their co-movement.